Understanding Past Weather to Improve Future Predictions
|Gilbert Compo, Jeffrey Whitaker, Prashant Sardeshmukh
Science Writer: Barb DeLuisi
Raw observations in hand, the next step is to combine the historical
observations based on physical principles while taking into account that the
data have errors and discrepancies, only some of which can be corrected.
Correctable errors in observed weather data can come from a variety of sources,
such as recording the wrong elevation where the observations were taken, or
errors in transcription, such as an observation of 1001 millibars of atmospheric
pressure being recorded as 1010. Other discrepancies include the difference two
barometers might have when separated by several kilometers.
(Right) Weather map of upper-troposphere made using only 308 surface pressure observations (blue dots).
Shows a close match to the map made using all available observations (left).
The researchers apply a numerical weather prediction model and a Kalman filter
to the data to combine the imperfect pressure observations in a process called
data assimilation. The filter step, named after Rudolph
E. Kalman, provides a mathematical way to create the weather map for a particular time by blending all of
the observations with a numerical weather model. This blending takes into account the
meteorological situation and the error in the observations. In the case of
creating the weather maps for a 100 years ago, very few observations are
available and those observations are only at the Earth's surface, but the data
assimilation procedure creates a complete map for the entire troposphere. "What
we have shown is that the map for the entire troposphere is very good even
though we have only used the surface pressure observations," says Compo. The
filter can change continuously based on the location on the globe, the number of
observations, or the meteorology. The filtering procedure currently takes one
day to process one month worth of global data. The resulting weather maps are
then given to the numerical weather prediction model to make a forecast of the
past weather. The quality of these forecasts of the past will help us know if
our weather maps are of a sufficient quality to be used for scientific study.